purge_addressof bootstrap failure (was fix opt/8634)

[Andreas Tobler]

Hi,

is the fix below ready to go in ? Or are there some other problems with?

As reported, the fix below makes the two tc's pass on darwin and linux-ppc.

Thanks,
Andreas

Zdenek Dvorak wrote:

>>These two ?
>>
>>FAIL: gcc.c-torture/compile/990107-1.c,  -O1

>>FAIL: gcc.c-torture/compile/990203-1.c,  -O1

> here's the fix.
> 
> Zdenek
> 
> 	* function.c (purge_addressof): Re-process notes after postponed
> 	insns processing.
> 
> Index: function.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/function.c,v
> retrieving revision 1.415
> diff -c -3 -p -r1.415 function.c
> *** function.c	9 Apr 2003 06:32:21 -0000	1.415
> --- function.c	10 Apr 2003 11:59:29 -0000
> *************** purge_addressof (insns)
> *** 3407,3412 ****
> --- 3407,3413 ----
>   {
>     rtx insn, tmp;
>     htab_t ht;
> +   rtx postponed_notes = NULL_RTX;
>   
>     /* When we actually purge ADDRESSOFs, we turn REGs into MEMs.  That
>        requires a fixup pass over the instruction stream to correct
> *************** purge_addressof (insns)
> *** 3431,3451 ****
>   
>   	if (! purge_addressof_1 (&REG_NOTES (insn), NULL_RTX, 0, 0, 0, ht))
>   	  {
> ! 	    /* If we could not replace the ADDRESSOFs in the insn's notes,
> ! 	       we can just remove the offending notes instead.  */
> ! 	    rtx note;
> ! 
> ! 	    for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
> ! 	      {
> ! 		/* If we find a REG_RETVAL note then the insn is a libcall.
> ! 		   Such insns must have REG_EQUAL notes as well, in order
> ! 		   for later passes of the compiler to work.  So it is not
> ! 		   safe to delete the notes here, and instead we abort.  */
> ! 		if (REG_NOTE_KIND (note) == REG_RETVAL)
> ! 		  abort ();
> ! 		if (for_each_rtx (&note, is_addressof, NULL))
> ! 		  remove_note (insn, note);
> ! 	      }
>   	  }
>         }
>   
> --- 3432,3440 ----
>   
>   	if (! purge_addressof_1 (&REG_NOTES (insn), NULL_RTX, 0, 0, 0, ht))
>   	  {
> ! 	    /* The failure may be due to postponed insns; so let's try this
> ! 	       again after the postponed insns are processed.  */
> ! 	    postponed_notes = alloc_INSN_LIST (insn, postponed_notes);
>   	  }
>         }
>   
> *************** purge_addressof (insns)
> *** 3460,3465 ****
> --- 3449,3482 ----
>         if (! purge_addressof_1 (&PATTERN (insn), insn,
>   			       asm_noperands (PATTERN (insn)) > 0, 0, 0, ht))
>   	abort ();
> +     }
> + 
> +   /* And postponed notes.  */
> +   while (postponed_notes)
> +     {
> +       insn = XEXP (postponed_notes, 0);
> +       tmp = postponed_notes;
> +       postponed_notes = XEXP (postponed_notes, 1);
> +       free_INSN_LIST_node (tmp);
> + 
> +       if (! purge_addressof_1 (&REG_NOTES (insn), NULL_RTX, 0, 0, 0, ht))
> + 	{
> + 	  /* If we could not replace the ADDRESSOFs in the insn's notes,
> + 	     we can just remove the offending notes instead.  */
> + 	  rtx note;
> + 
> + 	  for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
> + 	    {
> + 	      /* If we find a REG_RETVAL note then the insn is a libcall.
> + 		 Such insns must have REG_EQUAL notes as well, in order
> + 		 for later passes of the compiler to work.  So it is not
> + 		 safe to delete the notes here, and instead we abort.  */
> + 	      if (REG_NOTE_KIND (note) == REG_RETVAL)
> + 		abort ();
> + 	      if (for_each_rtx (&note, is_addressof, NULL))
> + 		remove_note (insn, note);
> + 	    }
> + 	}
>       }
>   
>     /* Clean up.  */
>